		FeedBack \| Contact Us \| Links \| Site Map

Welcome to Journal of Theoretical and Applied Information Technology

Submit Paper / Call for Papers

Journal receives papers in continuous flow and we will consider articles from a wide range of Information Technology disciplines encompassing the most basic research to the most innovative technologies. Please submit your papers electronically to our submission system at http://jatit.org/submit_paper.php in an MSWord, Pdf or compatible format so that they may be evaluated for publication in the upcoming issue. This journal uses a blinded review process; please remember to include all your personal identifiable information in the manuscript before submitting it for review, we will edit the necessary information at our side. Submissions to JATIT should be full research / review papers (properly indicated below main title).

Journal of Theoretical and Applied Information Technology

August 2024 | Vol. 102 No.16

Title:	A NEW CHANNEL ESTIMATION FRAMEWORK FOR 6G INTELLIGENT REFLECTING SURFACE-ENABLED MIMO
Author:	NAZIYA BEGUM, DR. MUHIEDDIN AMER, DR. OMAR ABDUL LATIF
Abstract:	The Intelligent Reflecting Surface (IRS) serves as a technology enabling passive manipulation of wave properties like amplitude, frequency, phase, and polarization through reflection. This technology is poised to revolutionize wireless communication by enhancing spectrum and energy efficiency while demanding minimal energy consumption. However, in scenarios where an IRS assists a base station (BS) with multiple antennas and user equipment (UE) with a single antenna, obtaining Instantaneous Channel State Information (I-CSI) for every link at the IRS poses challenges due to the numerous reflective elements and passive operation of the IRS. This imposes additional burdens on the system, necessitating the integration of radio-frequency chains into the IRS system. To address this, the paper proposes a three-phase pilot-based channel estimation framework for uplink multiuser communications, utilizing modular redundancy to reduce the time needed for channel estimation. The framework leverages IRS to achieve this goal by estimating the direct channels between UEs and BSs, as well as the reflected channels between a single UEs, IRS, and BS.
Keywords:	6G, IRS, MIMO, Channel estimation, Uplink, Downlink, Power utilization
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	USING DECISION TREE TECHNIQUE TO ANALYZE STUDENTS’ REFLECTIVE THINKING, FEEDBACK, AND PERFORMANCE
Author:	SITI KHADIJAH MOHAMAD, ZAIDATUN TASIR, SITI NAZLEEN ABDUL RABU
Abstract:	This study aims to apply a decision tree technique to understand how learning performance patterns in an educational blogging environment can be formulated and forecasted based on reflective thinking skills, feedback types, and performance test increment levels. A case study research design was adopted, involving qualitative data from students’ reflections on blogs and quantitative data on students’ performance. Data collection spanned 14 weeks and involved 18 postgraduate students enrolled in the Authoring System course. The data was prepared in .arff format and mined using the WEKA 3.6.6 machine learning toolkit to generate learning performance patterns. A random tree algorithm was applied to classify the data and evaluated using a three-fold cross-validation parameter. As a result, eight learning performance patterns were generated for three increment categories: P3, P4, and P5. From the generated patterns, it is evident that the higher the increment category of learning performance, the higher the levels of reflective thinking skills and feedback types utilised by students in their reflections. Moreover, Descriptive Reflection (DR) was noted as the most influential attribute differentiating the variables predicting all three learning increment classes. In addition to learning performance patterns, the overall performance of this predictive model was moderate (recall = 50%, precision = 50%) and acceptable (ROC = 61%). The findings are beneficial in identifying at-risk students, such as those struggling with reflection and displaying lower reflective thinking skills and feedback. This can alert instructors to take early action for intervention purposes.
Keywords:	Reflective Thinking, Feedback, Learning Performance, Decision Tree Data Mining, Blogging
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	A CONCEPTUAL FRAMEWORK FOR LEVERAGING CLOUD AND FOG COMPUTING IN DIABETES PREDICTION VIA MACHINE LEARNING ALGORITHMS: A PROPOSED IMPLEMENTATION
Author:	EDMIRA XHAFERRA, FLORIJE ISMAILI2, ELDA CINA, ANILA MITRE
Abstract:	This paper presents a theoretical framework for forecasting diabetes in Albania by combining cloud and fog computing with machine learning methods. The framework is designed to improve the effectiveness, expandability, and ethical accountability of diabetes management systems. It is customized to address the unique difficulties faced in the Albanian healthcare setting, such as restricted data accessibility and infrastructure limitations. The research used a mixed-methods approach, integrating both quantitative and qualitative procedures. In terms of quantity, it uses several data sources such as Electronic Health Records (EHRs), wearable devices, and laboratory testing. These data sources are subjected to thorough data preparation and feature selection algorithms. Machine learning models are assessed by employing measures such as accuracy and recall in conjunction with cross-validation procedures. The framework’s practicality and usability are evaluated using interviews, focus groups, and observational studies conducted in clinical settings to provide a qualitative assessment. This research is noteworthy due to its pioneering combination of cloud and fog computing with data analytics in the healthcare field, specifically in predicting diabetes. It creates new knowledge in integrating machine learning with healthcare data analytics to develop scalable and efficient predictive models for low-resource settings. It focuses on the requirement for predictive models that can operate in real-time and scale well, especially in contexts with limited resources, such as Albania. The study focuses on patient privacy, data security, and equal access to health services, making a valuable contribution to the academic discussion on the ethical implementation of AI in healthcare. This methodology not only enhances digital health research but also establishes a precedent for customizing technology solutions to address unique regional healthcare concerns.
Keywords:	Diabetes, Prediction, machine learning, cloud computing, fog computing
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	ADVANCING HISTOPATHOLOGIC CANCER DETECTION USING DIVERSE CNN ARCHITECTURES AND TRANSFER LEARNING
Author:	N.NAGA SUBRAHMANYESWARI, M.H.M.KRISHNA PRASAD
Abstract:	Histopathologic cancer detection plays a crucial role in early diagnosis and treatment planning. This research explores the application of distinct convolutional neural network (CNN) architectures, including VGG, InceptionV3, ResNet50, DenseNet121, and a custom CNN, for improving histopathologic cancer detection through transfer learning and advanced data augmentation techniques. Cancer Detection dataset is collected from Kaggle serves as the foundation for experiments. Transfer learning is employed from general image datasets like ImageNet and medical imaging datasets, tailoring the models to histopathologic characteristics. Each CNN architecture is examined independently to understand its unique contribution to feature extraction. Advanced data augmentation strategies, carefully designed to address limited annotated data, enhance the generalization capabilities of each model. These strategies, including rotation, shift, shear, and zoom, contribute to improved performance in histopathologic cancer detection. The work presents comprehensive analyses of each model's performance metrics, providing insights into their strengths in capturing intricate histological patterns. Results showcase the effectiveness of the individual CNN architectures, each demonstrating superior performance metrics. The findings underscore the significance of selecting appropriate architectures for specific tasks, contributing to advancements in automated histopathologic analysis with potential applications in early cancer diagnosis and treatment planning.
Keywords:	Histopathologic Cancer, CNN, Transfer Learning, VGG, Resnet, Inception
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	EMERGING AMBIDEXTROUS OPPORTUNITIES: HOW MALAYSIAN GLCS CAN LEVERAGE ARTIFICIAL INTELLIGENCE
Author:	K.INDRAVATHY, NOORLIZAWATI ABD RAHIM
Abstract:	Government-linked companies (GLCs) are integral to Malaysia's economic growth, operating under substantial government ownership and control across various sectors. In the digital era, it is crucial for GLCs to adopt advanced technologies, particularly artificial intelligence (AI), to enhance their performance and remain competitive against non-GLCs. This study investigates the role of AI in GLCs to provide ambidextrous opportunities, addressing the research gap focused on AI adoption and potential impact in Malaysian GLCs. A Systematic Literature Review (SLR) was conducted using the PRISMA method to analyze 48 peer-reviewed articles from the Web of Science (WOS) and Scopus databases, published between 2013 and 2024. The results revealed various roles of AI that can be leveraged by GLCs as ambidextrous opportunities, including automating financial tasks and services, enhancing transparency in procurement, optimizing supply chain resources, improving public administration, strengthening policy governance, optimizing marketing efforts, enhancing human resource management, boosting corporate readiness, optimizing energy management, managing sustainable natural resources, revolutionizing healthcare, implementing smart farming, enhancing e-commerce strategies, optimizing renewable energy utilization, enhancing smart grid management, increasing organizational agility, developing managerial skills, enhancing product development, and improving manufacturing efficiency. The study provides managerial recommendations for integrating AI across various sectors, aiming to boost competitiveness, operational efficiency, and innovation. This research contributes to the literature by offering practical insights and strategic guidance for policymakers and managers in leveraging AI to create ambidextrous opportunities, ensuring sustained growth and competitiveness in the digital age. This study recommends future research focus on the issues and challenges of leveraging AI in GLCs.
Keywords:	Artificial intelligence, government-linked companies, ambidextrous opportunities, digital transformation, adoption, systematic literature review, technology
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	EDGE ENHANCE SPARSE DEEP AUTO ENCODER MODEL FOR HIGH-ACCURACY BRAIN TUMOR DETECTION IN MRI IMAGES
Author:	SATYAVATI JAGA, K. RAMA DEVI
Abstract:	Brain tumors are among the most lethal cancerous diseases, with their severity making them a leading cause of cancer-related deaths. The treatment of brain tumors depends on the tumor's type, location, and size. Solely relying on human inspection for accurate categorization can result in potentially dangerous situations. This manual diagnostic process can be enhanced and expedited through the use of an automated Computer Aided Diagnosis (CAD) system. This research embarks on a ground-breaking path, leveraging cutting-edge deep learning methods to revolutionize anomaly detection and classification of brain tumor in MRI images. The aim of this work is to address the crucial area of brain tumor detection by presenting a new and innovative methodology that surpasses the limitations of current techniques. Central to this investigation is the Edge Enhance Sparse Deep Auto Encoder (EES-DAE) model, which introduces a network designed to detect and enhance edge saliency through deep learning (DL) techniques. The significance of the EES-DAE model is highlighted by its multidimensional approach, which greatly enhances the detection of brain tumors. The process starts with a pre-processing phase where a Wiener filter is applied to enhance the quality of brain MRI scans, providing a robust basis for further analysis. Next, a pixel normalization elimination step is performed to extract essential features while minimizing the effects of noise interference. Additionally, the intermediate stages of the methodology smoothly incorporate the 2D Discrete Cosine Transform (DCT) and entropy filters, orchestrating the precise extraction of complex features that greatly enhance the model's accuracy. The integration of the Edge Enhance Sparse Deep Auto Encoder (EES-DAE) model with the capabilities of soft-max entropy classification marks the apex of this innovative journey. The BRATS MRI brain images are utilized to assess the effectiveness of the developed techniques. This seamless fusion of advanced techniques achieves a remarkable outcome: the detection of brain tumor regions with an impressive accuracy rate of 99.13%.
Keywords:	MRI, CAD System, EES-DAE Model, DL, 2D-DCT, CNN, Softmax, Preprocessing, Wiener Filter, Pixel Normalization Elimination, Feature Extraction.
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	PERFORMANCE COMPARISON OF REINFORCEMENT LEARNING ALGORITHMS IN THE CARTPOLE GAME USING UNITY ML-AGENTS
Author:	EUN-HYEONG JO, YOUNGSIK KIM
Abstract:	Reinforcement learning is a field of machine learning where agents learn optimal actions through trial and error interactions with their environment. Games provide an effective benchmark to evaluate and compare the performance of reinforcement learning algorithms. This study utilized Unity's ML-Agents to implement the 'CartPole' game and applied various algorithms, including the Deep Q-Network (DQN), Advantage Actor-Critic (A2C), and Proximal Policy Optimization (PPO), to compare their performance. The primary research contribution of this work is the systematic comparison of these algorithms within a consistent environment, providing insights into their respective strengths and weaknesses. The study presents detailed analyses of the learning processes and outcomes of each algorithm, highlighting the DQN's superior performance in terms of stability and efficiency. Additionally, this work contributes new knowledge by demonstrating the practical applications and potential of reinforcement learning algorithms in simple game environments, thereby informing future developments in more complex domains.
Keywords:	Reinforcement Learning Algorithm, Deep Q-Network (DQN) Algorithm, Performance Comparison, CartPole Game
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	MODELING OF SWITCHING PROPERTIES OF SEMICONDUCTOR OPTICAL AMPLIFIER GATE BY THE XGM MECHANISM
Author:	A. ELYAMANI, A. MOUMEN, H.BOUSSETA, A. ZATNI
Abstract:	The semiconductor optical amplifiers (SOAs) are multifunctional optoelectronic components. The design of SOAs necessarily needs to go through a phase of modelling which allows studying a theoretical characterization of the SOA and anticipating its reactions on the basis of the operating conditions. This work presents a depth theoretical analysis of the wavelength conversion dynamics by the cross-gain modulation (XGM) in SOA. For this we will use a dynamic algorithm that we have already developed and which will allow us the efficient implementation of the model by means of the finite difference method (FDM), carrier density rate equation and a set of travelling-wave equations. This numerical model enables us to describe the propagation of a rectangular optical signal passing through the SOA and takes into account the effect of the nonlinearity by realizing a pump-probe assemblage to control the power and state of polarization at the entering of the SOA. The model can be used to investigate the effects of material and geometrical parameters on the switching speed properties, extinction ratio, recovery rate and output power of the signals in SOAs. Thus, we can then study the most important properties of SOA for the use in amplification, wavelength conversion, optical logic gates, code conversion and regeneration which demonstrates the versatility of our model.
Keywords:	Modeling, Cross-Gain Modulation (XGM), Semiconductor Optical Amplifier (SOA), Switching Speed, Extinction Ratio (ER), Recovery Rate, Finite Difference Method (FDM).
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	IMPROVING PORT SCAN CYBERSECURITY RISKS DETECTION USING FEATURES SELECTION TECHNIQUES WITH ML ALGORITHMS
Author:	RAMI SHEHAB, RANA ALRAWASHDEH, ROMEL AL-ALI, TAYSEER ALKHDOUR, MOHAMMED AMIN ALMAIAH
Abstract:	Malicious automated tools use port scan attacks to explore a target systems network ports aiming to find open ports and potential weaknesses. Port scanning can serve as a tool for system admins and cybersecurity experts to explore the main weaknesses in the network. Several countermeasures have been employed to defend against port attacks such as firewalls, intrusion detection systems (IDS) and network monitoring tools. These countermeasures aim to prevent malicious port scanning attacks. Based on that, port scan risks assessment is one of the essential step for detecting threats, vulnerabilities and protect the network security. Thus, this work aims to detect port scan attacks using features selection techniques with machine learning (ML) algorithms to reduce cyber-attacks being successful. To achieve this objective, we used three feature selection methods namely, ant colony algorithm (ACO), genetic algorithm (GA), and gray wolf optimization (GWO) with machine learning algorithms such as support vector machine (SVM), and nearest neighbor (KNN). The proposed work has been evaluated using confusion matrix measurements in terms of precision, recall, F1 score and accuracy. The study findings show that the percentage of risks and attacks detection over 99% for all proposed models. This study confirms that through the use of feature selection algorithms and machine learning methods, can help researchers to identify port behaviors and attacks in more efficiently.
Keywords:	Port Scan Attacks; Cyber-Risk; Machine Learning; Feature Selections; ACO; GA; GWO; SVM; KNN.
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	A FRAMEWORK FOR ARTIFICIAL INTELLIGENCE RISK MANAGEMENT
Author:	DAVID LAU KEAT JIN, GANTHAN NARAYANA SAMY, FIZA ABDUL RAHIM, NURAZEAN MAAROP, MAHISWARAN SELVANANTHAN, MAZLAN ALI, VALLIAPPAN RAMAN
Abstract:	Artificial Intelligence (AI) affords tremendous benefits to multiple sectors and businesses as its capabilities extend to different domain of activities. Notwithstanding the benefits that it brings, there are also potential risks which cause concerns by its users and those impacted by its use. Effective risk management is thus essential for organizations planning to deploy AI in high-risk applications. This study introduced a framework developed using a knowledge graph that stores and manages information on risk management, the AI life cycle, and stakeholder involvement, adhering to established standards. The framework facilitated the retrieval and generation of insights that support decision-making related to risk management, as it can represent interrelationships between entities more effectively than relational databases or typographies. The insights that can be generated include distribution of risks according to AI life cycle phases, the countermeasure that could treat the greatest number of risks and the countermeasure that produced the greatest change in terms of impact and probability to the identified risk. In this study, Cypher language was used to develop the framework, while Python language was used to generate the insights from the framework. Future studies may consider the integration of the framework in an enhanced Enterprise Risk Management framework to enable real-time update of related information and response by the organization.
Keywords:	Artificial Intelligence, Risk Management, AI Life Cycle, Stakeholder
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	ENHANCING CLOUD DATA SECURITY THROUGH LONG-TERM SECRET SHARING SCHEMES
Author:	SARA IBN EL AHRACHE, HASSAN BADIR
Abstract:	Cloud computing has experienced significant growth in recent years, becoming a cornerstone of modern IT infrastructure. Promising "infinite scalability and unlimited resources," cloud service providers offer on-demand access that often obscures the underlying computing infrastructure. The inherent complexity of virtualized, multi-tenant cloud environments surpasses that of traditional data centers, complicating service management, particularly in terms of security. Despite these challenges, the appealing features of cloud computing have led many organizations to adopt cloud storage services for their critical data. Users can store data remotely in the cloud and access it via thin clients when needed. However, data security remains a paramount concern due to the internet-based nature of cloud services, which limits user control over stored data. This paper proposes an innovative approach to enhance data security in cloud environments through a Long-Term Secret Sharing Scheme (SSS-LT). Secret sharing schemes partition and distribute data across multiple cloud service providers, thereby increasing data privacy and availability. Our proposed SSS-LT addresses a key limitation of existing secret sharing methods: the degradation of computational performance with large data sets. We conduct a theoretical analysis of the security and complexity factors influencing our approach and validate its efficacy through experimental evaluation, demonstrating its superiority over existing methods.
Keywords:	Cloud Computing, Data Security, Secret Sharing, Map Reduce
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	A SMART FAMEWORK TO GUIDE THE CUSTOMERS TO RAISE THE RETURN ON INVESTMENT
Author:	YOUSR AHMED ALAA ELDEEN, SHERIF ADEL ABD EL ALEEM, AHMED IBRAHIM BAHGAT EL SEDDAW
Abstract:	This study assesses the performance of various machine learning models and traditional statistical tools in forecasting the prices of key commodities, including gold, silver, crude oil, Brent oil, natural gas, and copper. The models evaluated encompass Random Forest Regression, Gradient Boosting Regression, Support Vector Regression, XGBoost, Long Short-Term Memory (LSTM), and Artificial Neural Networks (ANN), in addition to traditional econometric tools like SPSS and EViews. Performance metrics are based on the Root Mean Square Error (RMSE) statistic. The findings reveal that LSTM outperforms other models in capturing intricate time series patterns, particularly for copper price predictions, demonstrated by significantly lower RMSE values. While classical statistical methods and other machine learning models achieve reasonable accuracy, LSTM and ANN consistently show superior performance. Furthermore, an integrated model combining SPSS, EViews, and LSTM projections identifies top investment prospects. Gold and silver are highlighted as solid, safe-haven assets with highly accurate forecasts. Natural gas is noted for its precise price predictions and potential for substantial price increases. Copper stands out due to its excellent predictive accuracy and its capability to provide early warnings of gold price changes. The integration of traditional statistical approaches with advanced machine learning models offers a comprehensive framework for forecasting commodity prices and pinpointing the most promising investment opportunities in the commodities market.
Keywords:	Forecasting Commodity Prices, Machine Learning, , Long Short – Term Memory (LSTM), Statistical Package for the Social Sciences (SPSS), Econometric Views )EViews( .
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	OBJECT-ASPECT ORIENTED MODELS TO PETRI NETS MODEL, AN APPROACH FOR THE TRANSFORMATION, ANALYSIS AND VERIFICATION OF SOFTWARE SYSTEMS
Author:	MOUNA AOUAG, NOUHAD MERABET, WIAM KENNOUCHE
Abstract:	Object-Oriented Modeling (OOM) is a software design approach that structures and organizes code to accurately reflect reality, thereby facilitating its maintenance and evolution. However, it does have limitations in managing crosscutting concerns. Aspect-Oriented Modeling (AOM) provides solutions to these challenges, despite its lack of formal semantics. This underscores the importance of formal modeling, which does offer rigorous semantics. In this paper, we propose an approach to transform a detailed object-oriented sequence diagram into a detailed aspect-oriented sequence diagram, based on graph transformation. Subsequently, we propose a method to transform the aspect-oriented diagram into a Petri net. Our work begins with a single meta-modeling for the first approach, using graph grammar rules to achieve an aspect-oriented model. Then, we apply the second approach to the result of the first one, using two meta-modeling and graph grammar rules, resulting in a Petri net. We use the AToMPM modeling tool. Finally, we perform a property analysis with the TINA tool.
Keywords:	Object-Oriented Modeling, Aspect-Oriented Modeling, Petri nets, AToMPM, TINA.
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	DEEP LEARNING-DRIVEN CAR PARKING SPACE DETECTION: A YOLOR APPROACH
Author:	SHARFINA FAZA, ROMI FADILLAH RAHMAT, RINA ANUGRAHWATY, JESSICA WONG, FARHAD NADI
Abstract:	Parking is a critical facility that must be widely accessible, especially in public places such as tourist attractions, shopping centers, and offices. However, people are often delayed due to difficulty finding empty parking spaces, particularly during peak hours or when there are many visitors. This is partly because visitors do not know the exact location of available slots and must repeatedly circle the area to find one. Therefore, there is a pressing need to develop more efficient methods for detecting parking space availability to expedite and simplify the search process. To address this challenge, our study employs the You Only Learn One Representation (YOLOR) method to help detect available parking slots in three distinct locations: the underground parking area at 'Centre Point' shopping mall in Medan City, the university library parking lot, and the campus parking facility at the University of Sumatera Utara. YOLOR is one of the deep learning methods that has shown promising results in object detection tasks, making it a suitable choice for this parking slot identification problem. Based on our system's tests, we achieved accuracies of 95.71%, 95.74%, and 83.33% for each parking lot, respectively. For the occupied class, we obtained precisions of 0.99 and 0.97 in each parking lot, while for the empty class, we consistently achieved a precision of 1.00 across all lots. Recall rates were 0.96, 0.95, and 0.92 for the occupied class, and 0.95, 0.96, and 1.00 for the empty class in each parking lot. Our model was trained on a total of 3,180 images, comprising 11,666 empty slots and 11,843 occupied slots. Testing was conducted using six 30-minute to 1-hour long videos, each capturing various parking conditions under different weather (sunny, cloudy, or rainy) and lighting situations.
Keywords:	Parking, Parking Spaces, Detection, YOLOR, Deep Learning
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	IDENTIFICATION OF PAPAYA VARIETIES IN COMPUTER VISION AND DEEP LEARNING APPROACHES
Author:	SHARFINA FAZA, ROMI FADILLAH RAHMAT2, MERYATUL HUSNA, AJULIO PADLY SEMBIRING, ANJAS SUBHANUARI, FARHAD NADI
Abstract:	Papaya is one of the most popular tropical fruits and is widely cultivated in Indonesia. Papaya plants are easily found in various regions due to their good adaptation to the tropical climate. In addition to being delicious for consumption, papaya also offers many health benefits because of its high fiber content and benefits for the digestive system. There are various popular papaya varieties that are frequently consumed by the Indonesian people. However, the similarity in shape and skin color among these papaya varieties often makes it difficult to distinguish them, especially for laypeople. This poses a challenge in identifying papaya varieties, both for daily consumption purposes and in the marketing process. This research aims to develop a prototype that can identify papaya varieties using computer vision technology with the R-FCN ResNet101 algorithm. The data collection for this research was conducted through direct surveys at several markets that sell various papaya varieties. The data collected was in the form of 1000 papaya fruit images captured using a mobile phone camera. The collected data was then divided into training and testing datasets. The first step was image preprocessing, which consisted of resizing the images, labeling the training data, and converting it into TensorFlow Record files. The next step involved processing the data using computer vision technology with the R-FCN ResNet101 algorithm. The result of this step was a learned model, which was then used to analyze and identify the papaya varieties in the testing data. The test results achieved an accuracy of 97.5%.
Keywords:	Papaya, Varieties, Identify, Computer Vision, R-FCN ResNet101.
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	IMPROVING SEMANTIC SEGMENTATION OF MEDICAL IMAGES WITH FINE-TUNING TECHNIQUES
Author:	DR. K. SUDHA RANI, DR. A .SUMA LATHA, VAISHNAVI SATISH, PRADEEPA MEDISETTY
Abstract:	The study looks into ways to fine-tune semantic image segmentation challenges. We investigate how to improve existing models of neural networks for pixel-by-pixel image classification, with the goal of improving their specificity and accuracy. To improve the model's performance, we look into adapting pre-trained models to particular segmentation tasks. The parameters of our study are learning rate scheduling, optimizer, and data augmentation, all of which help to enhance the network’s segmentation performance. The outcomes of our experiments show how well the suggested fine-tuning techniques work to improve the specificity and generalizability of semantic segmentation models for medical images.
Keywords:	Deep Learning, Semantic Image Segmentation, Fine-Tuning, Model Optimization, Pre-trained Models, Data Augmentation, Learning Rate Scheduling, Image Processing, Neural Networks, Model Performance.
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	COMPARATIVE ANALYSIS ON THE PERFORMANCE OF NHPP-BASED SOFTWARE RELIABILITY MODEL FOLLOWING EXPONENTIAL LIFE DISTRIBUTION
Author:	SEUNG KYU PARK
Abstract:	In this study, the exponential-type life distribution, which is known to be suitable for reliability analysis of failure occurrence phenomena, was applied to the NHPP-based software reliability model, and then the performance of the applied model was newly studied. For this purpose, the failure time data requested by the developer was used. In conclusion, first, as a result of evaluating MSE and R^2, which are the selection criteria for efficient models to determine the suitability of the proposed model, the proposed models showed a performance of over 85% and were all efficient. Second, in the performance analysis using m(t) and λ(t), which are attribute functions that have a significant impact on the performance of the reliability model, the Inverse-exponential model showed excellent performance. Third, as a result of testing future reliability, the reliability of the Inverse-exponential and Rayleigh models showed a high and stable trend over time. Therefore, as a result of evaluating performance attribute data (MSE and R^2, m(t), λ(t), R ̂(τ)), it was concluded that the Inverse-exponential model had the best performance. Thus, this study can present basic design data along with algorithmic solution techniques that can analyze and predict performance attribute data needed by developers during the early software development process.
Keywords:	Exponential-basic, Exponential-type, Exponential-power, Inverse-exponential, Rayleigh, Software Reliability.
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	OPTIMIZED DEEP LEARNING FRAMEWORK FOR BRAIN TUMOR DETECTION AND CLASSIFICATION USING HYBRID VISUAL GEOMETRY GROUP-16 WITH REDUCED WEIGHTS VIA BUTTERFLY OPTIMIZATION
Author:	RAMYA NIMMAGADDA, Dr. P. KALPANA DEVI
Abstract:	Finding and classifying brain tumors are important parts of medical image analysis that need advanced deep-learning methods and optimization algorithms. Recognizing the urgent need for accurate methods in brain tumor diagnosis, we present a comprehensive approach integrating various stages, including data preprocessing. In this preprocessing phase, we employ techniques like aspect ratio normalization and resizing to form a standardized dataset. By standardizing image dimensions, we aim to improve subsequent processes like feature extraction and segmentation, reducing potential distortions. The suggested model is made by using Convolutional Neural Networks (CNN) to find patterns and traits that make tumor and non-tumor areas different from each other. To overcome the intricate sections and fine textures during down-sampling, the proposed model is hybridized with U-Net architecture which gives accurate and robust results of 98%. Furthermore, the Dice coefficient is measured using Intersection Over Union (IOU) to ensure whether it is robust to class imbalance. This shows an intuitive interpretation, with higher values of 0.83 and 0.9 indicating strong and better segmentation performance. The model is further developed with VGG-16 to classify the tumor grades. In terms of accurately segmenting the tumor grades, the learnt relevant characteristics that are derived from the segmented tumor photos provide a 73% level of satisfaction. In order to overcome the complexity and over-fitting problems, the Butterfly Optimization algorithm is hybridized with VGG-16 which gives an enhanced output in classifying the grades. The proposed model outperforms other Machine Learning (ML) and Deep Learning (DL) methods in tumor and non-tumor identification and categorization with 99.99% accuracy. To further evaluate the suggested model's performance, mobility, and energy economy, it is also implemented in JETSON Orin hardware.
Keywords:	Deep Learning, Convolutional Neural Networks (CNNs), U-Net, VGG-16, Butterfly Optimization Algorithm.
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	COMPARISON OF LEXICON-BASED METHOD, MACHINE LEARNING AND CHATGPT ON SENTIMENT ANALYSIS OF BIG CAP AND SMALL CAP COMPANIES IN UNITED STATE INDEXES
Author:	IMRAN KAMIL , NATHAR SHAH
Abstract:	Sentiment analysis is a natural language processing (NLP) method that identifies the sentiment contained in a body of text. It has gained significant attention due to its potential applications in various domains, including finance, marketing, and public opinion monitoring. In the financial sector, sentiment analysis is essential for analyzing market trends, forecasting stock prices, and guiding investment choices. This research paper compares the performance of lexicon-based method, machine learning technique, and ChatGPT in sentiment analysis of big cap and small cap companies in United States indexes using Twitter data. The purpose of implementing ChatGPT is to identify the usefulness of this well-known tool that is currently flooding the social media scene. The results show that Random Forest achieved the highest accuracy overall with 83.6% on big cap and 78.8% on small cap. ChatGPT sentiment has an accuracy of 77.44% on big cap and 72.43% on small cap. Meanwhile the lowest performing method is the TextBlob which has an accuracy of 46.52% on big cap and 43.57% on small cap. Random Forest is able to understand the context of tweets and handle slang terms and phrases, while ChatGPT is still under development but has the potential to perform better in the future. There are many slang terms and phrases that are used in the stock market that are not included in the TextBlob dictionary. Therefore, the performance of TextBlob is the least performing method.
Keywords:	Sentiment Analysis, Random Forest, Lexicon, Textblob, Machine Learning, Chatgpt
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text

Title:	AN AUTOMATED MULTIMODAL HYBRID SYSTEM FOR WEB CONTENT FACT-CHECKING BASED ON BERT LANGUAGE MODEL AND CONVOLUTIONAL NEURAL NETWORK
Author:	C. VISHNU MOHAN, N. V. CHINNASAMY
Abstract:	Over the last decade, people have been widely using online platforms for sharing information and for understanding the news that has been happening around them. Classification of social media texts, tweets etc., are one of the emerging areas of research in today’s world, especially when it comes to information about political and entertainment sectors. However, there are certain challenges due to the fact that most commonly used Machine Learning techniques have not proven to be optimal, when considering both textual and image data for fake content detection. This study explores the efficacy of a hybrid deep learning architecture that leverages BERT for text representation along with Convolutional Neural Network (CNN) for classifying news as real or fake.
Keywords:	Machine Learning, classification, deep learning, BERT, CNN
Source:	Journal of Theoretical and Applied Information Technology 31^st August 2024 -- Vol. 102. No. 16-- 2024
Full Text